kernel/ptrace.c

   1 /*
   2  * linux/kernel/ptrace.c
   3  *
   4  * (C) Copyright 1999 Linus Torvalds
   5  *
   6  * Common interfaces for "ptrace()" which we do not want
   7  * to continually duplicate across every architecture.
   8  */
   9
  10 #include <linux/capability.h>
  11 #include <linux/module.h>
  12 #include <linux/sched.h>
  13 #include <linux/errno.h>
  14 #include <linux/mm.h>
  15 #include <linux/highmem.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/smp_lock.h>
  18 #include <linux/ptrace.h>
  19 #include <linux/security.h>
  20 #include <linux/signal.h>
  21 #include <linux/audit.h>
  22 #include <linux/pid_namespace.h>
  23 #include <linux/syscalls.h>
  24
  25 #include <asm/pgtable.h>
  26 #include <asm/uaccess.h>
  27
  28 /*
  29  * ptrace a task: make the debugger its new parent and
  30  * move it to the ptrace list.
  31  *
  32  * Must be called with the tasklist lock write-held.
  33  */
  34 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
  35 {
  36         BUG_ON(!list_empty(&child->ptrace_list));
  37         if (child->parent == new_parent)
  38                 return;
  39         list_add(&child->ptrace_list, &child->parent->ptrace_children);
  40         remove_parent(child);
  41         child->parent = new_parent;
  42         add_parent(child);
  43 }
  44
  45 /*
  46  * Turn a tracing stop into a normal stop now, since with no tracer there
  47  * would be no way to wake it up with SIGCONT or SIGKILL.  If there was a
  48  * signal sent that would resume the child, but didn't because it was in
  49  * TASK_TRACED, resume it now.
  50  * Requires that irqs be disabled.
  51  */
  52 void ptrace_untrace(struct task_struct *child)
  53 {
  54         spin_lock(&child->sighand->siglock);
  55         if (task_is_traced(child)) {
  56                 if (child->signal->flags & SIGNAL_STOP_STOPPED) {
  57                         __set_task_state(child, TASK_STOPPED);
  58                 } else {
  59                         signal_wake_up(child, 1);
  60                 }
  61         }
  62         spin_unlock(&child->sighand->siglock);
  63 }
  64
  65 /*
  66  * unptrace a task: move it back to its original parent and
  67  * remove it from the ptrace list.
  68  *
  69  * Must be called with the tasklist lock write-held.
  70  */
  71 void __ptrace_unlink(struct task_struct *child)
  72 {
  73         BUG_ON(!child->ptrace);
  74
  75         child->ptrace = 0;
  76         if (!list_empty(&child->ptrace_list)) {
  77                 list_del_init(&child->ptrace_list);
  78                 remove_parent(child);
  79                 child->parent = child->real_parent;
  80                 add_parent(child);
  81         }
  82
  83         if (task_is_traced(child))
  84                 ptrace_untrace(child);
  85 }
  86
  87 /*
  88  * Check that we have indeed attached to the thing..
  89  */
  90 int ptrace_check_attach(struct task_struct *child, int kill)
  91 {
  92         int ret = -ESRCH;
  93
  94         /*
  95          * We take the read lock around doing both checks to close a
  96          * possible race where someone else was tracing our child and
  97          * detached between these two checks.  After this locked check,
  98          * we are sure that this is our traced child and that can only
  99          * be changed by us so it's not changing right after this.
 100          */
 101         read_lock(&tasklist_lock);
 102         if ((child->ptrace & PT_PTRACED) && child->parent == current &&
 103             (!(child->ptrace & PT_ATTACHED) || child->real_parent != current)
 104             && child->signal != NULL) {
 105                 ret = 0;
 106                 spin_lock_irq(&child->sighand->siglock);
 107                 if (task_is_stopped(child))
 108                         child->state = TASK_TRACED;
 109                 else if (!task_is_traced(child) && !kill)
 110                         ret = -ESRCH;
 111                 spin_unlock_irq(&child->sighand->siglock);
 112         }
 113         read_unlock(&tasklist_lock);
 114
 115         if (!ret && !kill)
 116                 wait_task_inactive(child);
 117
 118         /* All systems go.. */
 119         return ret;
 120 }
 121
 122 int __ptrace_may_attach(struct task_struct *task)
 123 {
 124         /* May we inspect the given task?
 125          * This check is used both for attaching with ptrace
 126          * and for allowing access to sensitive information in /proc.
 127          *
 128          * ptrace_attach denies several cases that /proc allows
 129          * because setting up the necessary parent/child relationship
 130          * or halting the specified task is impossible.
 131          */
 132         int dumpable = 0;
 133         /* Don't let security modules deny introspection */
 134         if (task == current)
 135                 return 0;
 136         if (((current->uid != task->euid) ||
 137              (current->uid != task->suid) ||
 138              (current->uid != task->uid) ||
 139              (current->gid != task->egid) ||
 140              (current->gid != task->sgid) ||
 141              (current->gid != task->gid)) && !capable(CAP_SYS_PTRACE))
 142                 return -EPERM;
 143         smp_rmb();
 144         if (task->mm)
 145                 dumpable = get_dumpable(task->mm);
 146         if (!dumpable && !capable(CAP_SYS_PTRACE))
 147                 return -EPERM;
 148
 149         return security_ptrace(current, task);
 150 }
 151
 152 int ptrace_may_attach(struct task_struct *task)
 153 {
 154         int err;
 155         task_lock(task);
 156         err = __ptrace_may_attach(task);
 157         task_unlock(task);
 158         return !err;
 159 }
 160
 161 int ptrace_attach(struct task_struct *task)
 162 {
 163         int retval;
 164         unsigned long flags;
 165
 166         audit_ptrace(task);
 167
 168         retval = -EPERM;
 169         if (task->pid <= 1)
 170                 goto out;
 171         if (same_thread_group(task, current))
 172                 goto out;
 173
 174 repeat:
 175         /*
 176          * Nasty, nasty.
 177          *
 178          * We want to hold both the task-lock and the
 179          * tasklist_lock for writing at the same time.
 180          * But that's against the rules (tasklist_lock
 181          * is taken for reading by interrupts on other
 182          * cpu's that may have task_lock).
 183          */
 184         task_lock(task);
 185         if (!write_trylock_irqsave(&tasklist_lock, flags)) {
 186                 task_unlock(task);
 187                 do {
 188                         cpu_relax();
 189                 } while (!write_can_lock(&tasklist_lock));
 190                 goto repeat;
 191         }
 192
 193         if (!task->mm)
 194                 goto bad;
 195         /* the same process cannot be attached many times */
 196         if (task->ptrace & PT_PTRACED)
 197                 goto bad;
 198         retval = __ptrace_may_attach(task);
 199         if (retval)
 200                 goto bad;
 201
 202         /* Go */
 203         task->ptrace |= PT_PTRACED | ((task->real_parent != current)
 204                                       ? PT_ATTACHED : 0);
 205         if (capable(CAP_SYS_PTRACE))
 206                 task->ptrace |= PT_PTRACE_CAP;
 207
 208         __ptrace_link(task, current);
 209
 210         force_sig_specific(SIGSTOP, task);
 211
 212 bad:
 213         write_unlock_irqrestore(&tasklist_lock, flags);
 214         task_unlock(task);
 215 out:
 216         return retval;
 217 }
 218
 219 static inline void __ptrace_detach(struct task_struct *child, unsigned int data)
 220 {
 221         child->exit_code = data;
 222         /* .. re-parent .. */
 223         __ptrace_unlink(child);
 224         /* .. and wake it up. */
 225         if (child->exit_state != EXIT_ZOMBIE)
 226                 wake_up_process(child);
 227 }
 228
 229 int ptrace_detach(struct task_struct *child, unsigned int data)
 230 {
 231         if (!valid_signal(data))
 232                 return -EIO;
 233
 234         /* Architecture-specific hardware disable .. */
 235         ptrace_disable(child);
 236         clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 237
 238         write_lock_irq(&tasklist_lock);
 239         /* protect against de_thread()->release_task() */
 240         if (child->ptrace)
 241                 __ptrace_detach(child, data);
 242         write_unlock_irq(&tasklist_lock);
 243
 244         return 0;
 245 }
 246
 247 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
 248 {
 249         int copied = 0;
 250
 251         while (len > 0) {
 252                 char buf[128];
 253                 int this_len, retval;
 254
 255                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 256                 retval = access_process_vm(tsk, src, buf, this_len, 0);
 257                 if (!retval) {
 258                         if (copied)
 259                                 break;
 260                         return -EIO;
 261                 }
 262                 if (copy_to_user(dst, buf, retval))
 263                         return -EFAULT;
 264                 copied += retval;
 265                 src += retval;
 266                 dst += retval;
 267                 len -= retval;
 268         }
 269         return copied;
 270 }
 271
 272 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
 273 {
 274         int copied = 0;
 275
 276         while (len > 0) {
 277                 char buf[128];
 278                 int this_len, retval;
 279
 280                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 281                 if (copy_from_user(buf, src, this_len))
 282                         return -EFAULT;
 283                 retval = access_process_vm(tsk, dst, buf, this_len, 1);
 284                 if (!retval) {
 285                         if (copied)
 286                                 break;
 287                         return -EIO;
 288                 }
 289                 copied += retval;
 290                 src += retval;
 291                 dst += retval;
 292                 len -= retval;
 293         }
 294         return copied;
 295 }
 296
 297 static int ptrace_setoptions(struct task_struct *child, long data)
 298 {
 299         child->ptrace &= ~PT_TRACE_MASK;
 300
 301         if (data & PTRACE_O_TRACESYSGOOD)
 302                 child->ptrace |= PT_TRACESYSGOOD;
 303
 304         if (data & PTRACE_O_TRACEFORK)
 305                 child->ptrace |= PT_TRACE_FORK;
 306
 307         if (data & PTRACE_O_TRACEVFORK)
 308                 child->ptrace |= PT_TRACE_VFORK;
 309
 310         if (data & PTRACE_O_TRACECLONE)
 311                 child->ptrace |= PT_TRACE_CLONE;
 312
 313         if (data & PTRACE_O_TRACEEXEC)
 314                 child->ptrace |= PT_TRACE_EXEC;
 315
 316         if (data & PTRACE_O_TRACEVFORKDONE)
 317                 child->ptrace |= PT_TRACE_VFORK_DONE;
 318
 319         if (data & PTRACE_O_TRACEEXIT)
 320                 child->ptrace |= PT_TRACE_EXIT;
 321
 322         return (data & ~PTRACE_O_MASK) ? -EINVAL : 0;
 323 }
 324
 325 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t __user * data)
 326 {
 327         siginfo_t lastinfo;
 328         int error = -ESRCH;
 329
 330         read_lock(&tasklist_lock);
 331         if (likely(child->sighand != NULL)) {
 332                 error = -EINVAL;
 333                 spin_lock_irq(&child->sighand->siglock);
 334                 if (likely(child->last_siginfo != NULL)) {
 335                         lastinfo = *child->last_siginfo;
 336                         error = 0;
 337                 }
 338                 spin_unlock_irq(&child->sighand->siglock);
 339         }
 340         read_unlock(&tasklist_lock);
 341         if (!error)
 342                 return copy_siginfo_to_user(data, &lastinfo);
 343         return error;
 344 }
 345
 346 static int ptrace_setsiginfo(struct task_struct *child, siginfo_t __user * data)
 347 {
 348         siginfo_t newinfo;
 349         int error = -ESRCH;
 350
 351         if (copy_from_user(&newinfo, data, sizeof (siginfo_t)))
 352                 return -EFAULT;
 353
 354         read_lock(&tasklist_lock);
 355         if (likely(child->sighand != NULL)) {
 356                 error = -EINVAL;
 357                 spin_lock_irq(&child->sighand->siglock);
 358                 if (likely(child->last_siginfo != NULL)) {
 359                         *child->last_siginfo = newinfo;
 360                         error = 0;
 361                 }
 362                 spin_unlock_irq(&child->sighand->siglock);
 363         }
 364         read_unlock(&tasklist_lock);
 365         return error;
 366 }
 367
 368
 369 #ifdef PTRACE_SINGLESTEP
 370 #define is_singlestep(request)          ((request) == PTRACE_SINGLESTEP)
 371 #else
 372 #define is_singlestep(request)          0
 373 #endif
 374
 375 #ifdef PTRACE_SINGLEBLOCK
 376 #define is_singleblock(request)         ((request) == PTRACE_SINGLEBLOCK)
 377 #else
 378 #define is_singleblock(request)         0
 379 #endif
 380
 381 #ifdef PTRACE_SYSEMU
 382 #define is_sysemu_singlestep(request)   ((request) == PTRACE_SYSEMU_SINGLESTEP)
 383 #else
 384 #define is_sysemu_singlestep(request)   0
 385 #endif
 386
 387 static int ptrace_resume(struct task_struct *child, long request, long data)
 388 {
 389         if (!valid_signal(data))
 390                 return -EIO;
 391
 392         if (request == PTRACE_SYSCALL)
 393                 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 394         else
 395                 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 396
 397 #ifdef TIF_SYSCALL_EMU
 398         if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
 399                 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 400         else
 401                 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 402 #endif
 403
 404         if (is_singleblock(request)) {
 405                 if (unlikely(!arch_has_block_step()))
 406                         return -EIO;
 407                 user_enable_block_step(child);
 408         } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
 409                 if (unlikely(!arch_has_single_step()))
 410                         return -EIO;
 411                 user_enable_single_step(child);
 412         }
 413         else
 414                 user_disable_single_step(child);
 415
 416         child->exit_code = data;
 417         wake_up_process(child);
 418
 419         return 0;
 420 }
 421
 422 int ptrace_request(struct task_struct *child, long request,
 423                    long addr, long data)
 424 {
 425         int ret = -EIO;
 426
 427         switch (request) {
 428         case PTRACE_PEEKTEXT:
 429         case PTRACE_PEEKDATA:
 430                 return generic_ptrace_peekdata(child, addr, data);
 431         case PTRACE_POKETEXT:
 432         case PTRACE_POKEDATA:
 433                 return generic_ptrace_pokedata(child, addr, data);
 434
 435 #ifdef PTRACE_OLDSETOPTIONS
 436         case PTRACE_OLDSETOPTIONS:
 437 #endif
 438         case PTRACE_SETOPTIONS:
 439                 ret = ptrace_setoptions(child, data);
 440                 break;
 441         case PTRACE_GETEVENTMSG:
 442                 ret = put_user(child->ptrace_message, (unsigned long __user *) data);
 443                 break;
 444         case PTRACE_GETSIGINFO:
 445                 ret = ptrace_getsiginfo(child, (siginfo_t __user *) data);
 446                 break;
 447         case PTRACE_SETSIGINFO:
 448                 ret = ptrace_setsiginfo(child, (siginfo_t __user *) data);
 449                 break;
 450         case PTRACE_DETACH:      /* detach a process that was attached. */
 451                 ret = ptrace_detach(child, data);
 452                 break;
 453
 454 #ifdef PTRACE_SINGLESTEP
 455         case PTRACE_SINGLESTEP:
 456 #endif
 457 #ifdef PTRACE_SINGLEBLOCK
 458         case PTRACE_SINGLEBLOCK:
 459 #endif
 460 #ifdef PTRACE_SYSEMU
 461         case PTRACE_SYSEMU:
 462         case PTRACE_SYSEMU_SINGLESTEP:
 463 #endif
 464         case PTRACE_SYSCALL:
 465         case PTRACE_CONT:
 466                 return ptrace_resume(child, request, data);
 467
 468         case PTRACE_KILL:
 469                 if (child->exit_state)  /* already dead */
 470                         return 0;
 471                 return ptrace_resume(child, request, SIGKILL);
 472
 473         default:
 474                 break;
 475         }
 476
 477         return ret;
 478 }
 479
 480 /**
 481  * ptrace_traceme  --  helper for PTRACE_TRACEME
 482  *
 483  * Performs checks and sets PT_PTRACED.
 484  * Should be used by all ptrace implementations for PTRACE_TRACEME.
 485  */
 486 int ptrace_traceme(void)
 487 {
 488         int ret = -EPERM;
 489
 490         /*
 491          * Are we already being traced?
 492          */
 493         task_lock(current);
 494         if (!(current->ptrace & PT_PTRACED)) {
 495                 ret = security_ptrace(current->parent, current);
 496                 /*
 497                  * Set the ptrace bit in the process ptrace flags.
 498                  */
 499                 if (!ret)
 500                         current->ptrace |= PT_PTRACED;
 501         }
 502         task_unlock(current);
 503         return ret;
 504 }
 505
 506 /**
 507  * ptrace_get_task_struct  --  grab a task struct reference for ptrace
 508  * @pid:       process id to grab a task_struct reference of
 509  *
 510  * This function is a helper for ptrace implementations.  It checks
 511  * permissions and then grabs a task struct for use of the actual
 512  * ptrace implementation.
 513  *
 514  * Returns the task_struct for @pid or an ERR_PTR() on failure.
 515  */
 516 struct task_struct *ptrace_get_task_struct(pid_t pid)
 517 {
 518         struct task_struct *child;
 519
 520         /*
 521          * Tracing init is not allowed.
 522          */
 523         if (pid == 1)
 524                 return ERR_PTR(-EPERM);
 525
 526         read_lock(&tasklist_lock);
 527         child = find_task_by_vpid(pid);
 528         if (child)
 529                 get_task_struct(child);
 530
 531         read_unlock(&tasklist_lock);
 532         if (!child)
 533                 return ERR_PTR(-ESRCH);
 534         return child;
 535 }
 536
 537 #ifndef arch_ptrace_attach
 538 #define arch_ptrace_attach(child)       do { } while (0)
 539 #endif
 540
 541 #ifndef __ARCH_SYS_PTRACE
 542 asmlinkage long sys_ptrace(long request, long pid, long addr, long data)
 543 {
 544         struct task_struct *child;
 545         long ret;
 546
 547         /*
 548          * This lock_kernel fixes a subtle race with suid exec
 549          */
 550         lock_kernel();
 551         if (request == PTRACE_TRACEME) {
 552                 ret = ptrace_traceme();
 553                 if (!ret)
 554                         arch_ptrace_attach(current);
 555                 goto out;
 556         }
 557
 558         child = ptrace_get_task_struct(pid);
 559         if (IS_ERR(child)) {
 560                 ret = PTR_ERR(child);
 561                 goto out;
 562         }
 563
 564         if (request == PTRACE_ATTACH) {
 565                 ret = ptrace_attach(child);
 566                 /*
 567                  * Some architectures need to do book-keeping after
 568                  * a ptrace attach.
 569                  */
 570                 if (!ret)
 571                         arch_ptrace_attach(child);
 572                 goto out_put_task_struct;
 573         }
 574
 575         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 576         if (ret < 0)
 577                 goto out_put_task_struct;
 578
 579         ret = arch_ptrace(child, request, addr, data);
 580         if (ret < 0)
 581                 goto out_put_task_struct;
 582
 583  out_put_task_struct:
 584         put_task_struct(child);
 585  out:
 586         unlock_kernel();
 587         return ret;
 588 }
 589 #endif /* __ARCH_SYS_PTRACE */
 590
 591 int generic_ptrace_peekdata(struct task_struct *tsk, long addr, long data)
 592 {
 593         unsigned long tmp;
 594         int copied;
 595
 596         copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), 0);
 597         if (copied != sizeof(tmp))
 598                 return -EIO;
 599         return put_user(tmp, (unsigned long __user *)data);
 600 }
 601
 602 int generic_ptrace_pokedata(struct task_struct *tsk, long addr, long data)
 603 {
 604         int copied;
 605
 606         copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
 607         return (copied == sizeof(data)) ? 0 : -EIO;
 608 }
 609
 610 #ifdef CONFIG_COMPAT
 611 #include <linux/compat.h>
 612
 613 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
 614                           compat_ulong_t addr, compat_ulong_t data)
 615 {
 616         compat_ulong_t __user *datap = compat_ptr(data);
 617         compat_ulong_t word;
 618         int ret;
 619
 620         switch (request) {
 621         case PTRACE_PEEKTEXT:
 622         case PTRACE_PEEKDATA:
 623                 ret = access_process_vm(child, addr, &word, sizeof(word), 0);
 624                 if (ret != sizeof(word))
 625                         ret = -EIO;
 626                 else
 627                         ret = put_user(word, datap);
 628                 break;
 629
 630         case PTRACE_POKETEXT:
 631         case PTRACE_POKEDATA:
 632                 ret = access_process_vm(child, addr, &data, sizeof(data), 1);
 633                 ret = (ret != sizeof(data) ? -EIO : 0);
 634                 break;
 635
 636         case PTRACE_GETEVENTMSG:
 637                 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
 638                 break;
 639
 640         default:
 641                 ret = ptrace_request(child, request, addr, data);
 642         }
 643
 644         return ret;
 645 }
 646
 647 #ifdef __ARCH_WANT_COMPAT_SYS_PTRACE
 648 asmlinkage long compat_sys_ptrace(compat_long_t request, compat_long_t pid,
 649                                   compat_long_t addr, compat_long_t data)
 650 {
 651         struct task_struct *child;
 652         long ret;
 653
 654         /*
 655          * This lock_kernel fixes a subtle race with suid exec
 656          */
 657         lock_kernel();
 658         if (request == PTRACE_TRACEME) {
 659                 ret = ptrace_traceme();
 660                 goto out;
 661         }
 662
 663         child = ptrace_get_task_struct(pid);
 664         if (IS_ERR(child)) {
 665                 ret = PTR_ERR(child);
 666                 goto out;
 667         }
 668
 669         if (request == PTRACE_ATTACH) {
 670                 ret = ptrace_attach(child);
 671                 /*
 672                  * Some architectures need to do book-keeping after
 673                  * a ptrace attach.
 674                  */
 675                 if (!ret)
 676                         arch_ptrace_attach(child);
 677                 goto out_put_task_struct;
 678         }
 679
 680         ret = ptrace_check_attach(child, request == PTRACE_KILL);
 681         if (!ret)
 682                 ret = compat_arch_ptrace(child, request, addr, data);
 683
 684  out_put_task_struct:
 685         put_task_struct(child);
 686  out:
 687         unlock_kernel();
 688         return ret;
 689 }
 690 #endif /* __ARCH_WANT_COMPAT_SYS_PTRACE */
 691
 692 #endif  /* CONFIG_COMPAT */